the autonomy of biological individuals and artificial models

A

sroocd©

K

1

eaooloiat1aa

o

0

BioSystems 91 (2008) 309–319

Available online at www.sciencedirect.com

The autonomy of biological individuals and artificial models

Alvaro Moreno ∗, Arantza Etxeberria, Jon UmerezDepartment of Logic and Philosophy of Science, University of the Basque Country

UPV-EHU Tolosa hirib.70//E-20018 Donostia, Spain

Received 7 March 2007; received in revised form 27 April 2007; accepted 11 May 2007

bstract

This paper aims to offer an overview of the meaning of autonomy for biological individuals and artificial models rooted in apecific perspective that pays attention to the historical and structural aspects of its origins and evolution. Taking autopoiesis and theecursivity characteristic of its circular logic as a starting point, we depart from some of its consequences to claim that the theory
f autonomy should also take into account historical and structural features. Autonomy should not be considered only in internalr constitutive terms, the largely neglected interactive aspects stemming from it should be equally addressed. Artificial modelsontribute to get a better understanding of the role of autonomy for life and the varieties of its organization and phenomenologicaliversity.
2007 Elsevier Ireland Ltd. All rights reserved.

egratio
eywords: Agency; Biological organization; Cognition; Function; Int
. Introduction

Autonomy means self-law, to be in charge. In gen-ral, this is understood either as the capacity to actccording to self-determined principles, or as the dutyf recognizing and respecting that aptitude about some-ne. In an ontological usage it may also mean that a givenevel or realm is relatively independent with respect tothers because it is ruled by its own norms. Yet, accord-ng to Maturana and Varela’s theory of autopoiesis,utonomous capacities stem from self-production andhey constitute an identity (Maturana and Varela, 1973,980, 1984). Then, not merely autonomous action, butlso autonomous being is the subject matter of this last
pproach.
The theory of autopoiesis placed the notion of auton-my at the center of the biological understanding of

∗ Corresponding author.E-mail address: [email protected] (A. Moreno).

303-2647/$ – see front matter © 2007 Elsevier Ireland Ltd. All rights reservdoi:10.1016/j.biosystems.2007.05.009

n; Self-organization

living beings in a moment (early seventies) when theatmosphere was probably more prepared for systemicdevelopments than inmediately before or after (Wimsatt,2007; Etxeberria & Umerez, 2006); (as it is well known,the biology of the second half of the 20th centuryrevolved around the concept of the gene to explainmost of living phenomenology, including developmentor evolution). Although for a long time the impact ofthis notion in main stream biology was not major, atpresent it has started to draw a lot more attention, espe-cially among those interested in a biology centered inthe organism. Examples of this are this early century’sreturn of Systems Biology (Boogerd et al., 2007; Kitano,2002; O’Malley and Dupre, 2005; Science, 2002), therenewed interest for the Kantian third Critique’s view oforganisms in the philosophy of biology (Van de Vijveret al., 2003), and the biologically inspired Artificial Life
and Robotics (Webb, 2001; Steels and Brooks, 1995;Beer, 1997; Nolfi and Floreano, 2000; Di Paolo, 2003).In all three (biology, philosophy and artificial science),the exploration of autonomy is likely to bring about
ed.

mailto:[email protected]

dx.doi.org/10.1016/j.biosystems.2007.05.009

System
310 A. Moreno et al. / Bio
new perspectives for research (Etxeberria et al., 2000).If Modernity1 thought that autonomy was a desiredconsequence of the human faculty of reason, contem-porary naturalism aims to understand life and cognitionas expressions of the autonomy of some material sys-tems. The focus is shifted from the actions deriving fromautonomy to the processes able to originate that capacity,to its conditions of possibility.

We want to stress here the significance of auton-omy for sciences aiming to understand living systems.In this sense, exploring autonomy should constitute aninevitable axis in the models, artificial systems, etc.involving phenomena such as organization, develop-ment, behavior or evolution. In our view, autonomy isthe main feature of life, the key notion for any attemptto define it.

Such tight a relation between life and autonomysuggests a complicated status for artificial autonomoussystems. On the one hand, if an artificial autonomous sys-tem would be produced, then we might call it “alive”. Thereason is that life may be understood in certain ways as acategory beyond the natural/artificial divide. Natural sys-tems generate spontaneously, whereas artificial systemsare created by an intelligent agent (generally human)2

through the use of complex cognitive capacities andresources. But, according to Hans Jonas (1966/2001)),life has some form of primacy for the living, in the sensethat it is recognized without the need to analyze it intoconstituent parts. Although none of the artificial systemsso far produced are alive, if a truly autonomous systemwould come into existence by means of human inter-vention, then its reality as autonomous or alive would
be more significant to us than the fact that it is artifi-cial; its artificiality would be somehow secondary withrespect to its aliveness. On the other hand, these sys-
1 The concept of autonomy has been used in many domains (politicalphilosophy, ethics, biology, robotics. . .), and it is not evident whetherthere are clearly traceable genealogical relations among the varioususages. In ancient Greece the term was applied to city-states andreferred to the political right of self-government. Later, modern phi-losophy extended the term to the self-determination of persons, bothpolitical and ethical. In the sense we claim for in philosophy of biology,artificial life and organismic robotics, the concept plays an importantrole in the definition of the identity and the interactive capacities ofindividuals.

2 It is possible to consider that artefacts or substances produced byanimals are equally artificial, and this view has the advantage of sit-uating human cognitive and technical abilities within an evolutionarycontinuum. It has the problem of blurring some intuitive differences,as both plastic and honey would appear to be equally artificial fromthis perspective, but we think that the distinction may be recovered if itwere needed to argue in a given context (for example, by distinguishingthe technological abilities of humans and other animals).

s 91 (2008) 309–319

tems would be also extensions of ourselves,3 effects ouragency has on the environment, even if their autonomymight prevent us from consider them as mere products(because they would be self-produced). It is doubtlessthat the exploration of autonomy poses a special kind ofchallenge for human agency, as it requires the reflex-ive form of epistemology searched by cyberneticians(Hayles, 1999).

However, we do not contend that (re)producing auton-omy should constitute a technological goal for researchon life. Concerning our understanding of life, whatreally matters is the construction of models of auton-omy (which, if material, are often called “artificialautonomous systems”). But modelling autonomy is acomplex task, a real challenge for an experimentalepistemology. Not being exhausted by the ontologicalproduction of autonomous system, it may be appealingas an activity in which the purposes and intervention ofmodellers interact with a dynamic and emergent system,with the result of understanding.

In that context, the goal of this paper is to offer anoverview of the meaning of autonomy for biologicalindividuals and artificial models rooted in a specific per-spective that pays attention to the historical and structuralaspects of its origins and evolution. In the next section wewill discuss the question of minimal autonomy, namely,what characterizes autonomy from less complex forms ofself-organization and self-maintenance. In Section 3 weanalyze the specific features of different levels, degreesand domains of autonomy. Finally, in Section 4 we con-sider the relation between autonomy and the artificial,mainly focusing on methodological and epistemologicalaspects.

2. Autopoiesis and minimal autonomy

Since the relation between the concepts of autonomyand life is so tight, we may wonder if, in the tran-sition between the non-living and the living there are(were) non-living organizations that can be consideredautonomous to some degree. Is the simplest biologicalform we know also the simplest form of autonomy?Are there reasons to think that non-organic or pre-biotic
forms of organization, simpler than known life, couldbe already autonomous? In other words, this is a ques-tion about self-organization, minimal autonomy and therelation between the two of them.
3 Keller (2007) proposes this understanding. She is inspired byTurner’s (2000) work on the way animals use their environmentalconstructions as extensions of their bodies.

System

itfeisit

obataa

siop

sihamnidtaiecesfdta

o2

A. Moreno et al. / Bio

Yet, it is essential to keep in mind that autopoiesiss an a-historical concept: it articulates an organiza-ion emerging from the dynamics of components, notrom evolution.4 As formulated by Maturana and Varela,volution will give rise to a manifold of structures deriv-ng from autopoiesis, but the organization remains theame. Thus, only an evolutionary account implying somencrease of complexity in autonomy may justify referringo minimal forms.

The concept of autopoiesis refers to a recursive netf component production that builds up its own physicalorder. The global net of component relations establishesself-maintaining dynamics, whose action brings about

he constitution of the system as an operational unit. Inutopoiesis, components and processes are entangled incyclic, recursive production logic.

Although the stress is made in the circular logic of theystem, and primarily it is an “internalist” perspective,n other places these authors also consider the relationf the system with its environment. For example, theyrovide the following definition in their glossary:

“Autonomy: the condition of subordinating allchanges to the maintenance of the organization. Self-asserting capacity of living systems to maintain theiridentity through the active compensation of deforma-tions”. (M&V 1980, p. 135).

This definition contains, at least, two aspects worthtressing in our attempt to establish the lower lim-ts at which autonomy may appear. One is that thereas to be an organization toward whose maintenancell changes must be subordinated. The other is thataintenance is active and self-asserting. Thus, phe-

omena like tornadoes, whirlpools and candle flames,n which self-organizing properties appear to a certainegree, are not autonomous. Indeed (some of) these sys-ems may have “homeostatic” capacities. For instance,

candle flame can compensate a small puff of air byncrease in temperature back to steady state and re-stablishment of vertical air flow from the flame. But weannot detect any form of active interaction or “agency”xerted by the system. In that sense, what distinguishesimple forms of self-organization and self-maintenancerom autonomy is, as we will explain later in more
etail, that the former merely react against external per-urbations, not being capable of displaying selectivections.
4 In this sense, Di Paolo (2005), notes that an autopoietic networkf processes reversed in time is still autopoietic. See also (Etxeberria,004).

s 91 (2008) 309–319 311

In a different view of autonomy, that of Pattee (i.e.,1972, 1973, 2001), self-organizing systems like torna-does are not entitled to be considered autonomous either,but for different reasons. His argument would be that anactive compensation requires that the system is capa-ble of some form of selection, as basic as it might be,between different alternatives. If no choice is available,no action can be exerted: the dynamics just results in theadvent of the next state of a temporal sequence deter-mined by laws. Is a planet, exerting its gravitational forceupon a given object, “active”? What about the water flowof a river eroding the surface of its bed? Or, to comecloser to our case, when a tree is bent by the wind andbounces back to its resting position, is it doing an “activecompensation” against the “deformation” induced bythe wind? According to Pattee, the three of them areexamples of systems governed by laws. However, theautonomy of the system depends on the existence ofinternal constraints able to channel its process dynamicsin one of the possible directions. This amounts to someform of additional causality with respect to law-governeddynamics, exerted by the system itself.

Thus, the existence of this type of causal con-straints in the system may demarcate the differencebetween self-organizing systems, as generally under-stood, and what people like Simon (1996) called“organized complexity”, using Weaver’s (1948) term.Usually, self-organizing systems are studied as a “one-shot, order-for-free” kind of process whose outcome isthe emergence of a global pattern starting from a uniformnon-linear dynamics below. This pattern is not able toproduce any internal selection within the system. How-ever, organized complexity involves some distinctionamong parts and their characteristic functions within thesystem. It often results from processes of composition orself-assembly between heterogeneous subsystems, withthe result of some form of hierarchy (as the mergingof autonomous identities in the origins of eukaryoticcells). Far from being an immediate process, this kindof organization can only result from iterative processesof self-organization over time (Keller, 2007).

The aim to make some progress concerning this essen-tial distinction is probably on the basis of Kauffman’s(2000) approach to autonomy, expressed in thermody-namic terms. Starting from the concept of “autocatalyticset”, the main condition required to consider a system asautonomous is that it is “able to perform at least one ther-modynamic work cycle” (Kauffman (2000), p. 4). This
capacity is implemented through a deep entanglementbetween work and constraints: “work begets constraintsbegets work”, as some form of closure in the abstractspace of catalytic tasks. This insight is based, on the

System

after some extensive exploration of self-organizing prop-erties in the last decades, it is clear that no natural notionof function emerges out of it. But it is also evident that


one hand, in Atkins’ view (1984) of work as a coordi-nated, coherent and constrained release of energy, and,on the other hand, in the recognition that work is abso-lutely necessary to build constraints. In other words, tobe autonomous, a system must do work; to be capableof work it requires constraints to channel the flow ofenergy in an appropriate way, and to build those con-straints the system requires appropriately constrainedenergy flows, that is to say, work. This is the circularityof the “work-constraint cycle”.

In our view, Kauffman’s account envisages how func-tionality or utility (implicit in the idea of work) can comeout of the causal circularity of the system, where thiscircularity is not only understood in terms of abstractrelations of component production, but also as an ener-getic logic sustaining the specific chemical recursivityof the system: all the processes ongoing in the sys-tem are constrained in order to satisfy the condition ofself-maintenance. This is how explanations in terms offunctions can and should be introduced on the studyof autonomous systems. However, unlike abstract com-putational functionalism and evolutionary etiologicalaccounts of function, a utilitarian notion of autonomyas differential contribution to self-maintenance (Col-lier, 1998; Christensen and Bickhard, 2002) uncoversthe deep entanglement between structures, functions andstability that is characteristic of autonomous systems.

Yet, in his work, like in autopoiesis, the notion ofrecursivity remains the conceptual nucleus of autonomyas identity preserving. The system is constituted as aseries of causal processes (energy transduction, com-ponent production, etc.) converging to a given (initial)state, as an indefinite repetition of the same loop. Thus,although these models admit variations, their essentialaspect is the existence of a pattern, which is maintainedthroughout the self-producing cycles, and constitutes the“identity” of the system.

Thus, from this perspective in order to preserve thesystem’s identity, its actions must counteract possibleperturbations from the environment, and further actionsof autonomous systems must always have that inter-nal reference only. As a consequence, the actions of anautonomous system in the environment are only sideeffects of the internal self-production. Although it issound to consider that the reference of autonomousagency is internal, that is to say, that the self-maintenancewill define what is relevant, a broad consideration ofagency suggests that the transformation of the envi-
ronment should be more than a passive consequenceof self-production: autonomous agents simultaneouslypreserve the self-produced identity and transform theenvironment, not only to make their more suitable to its
s 91 (2008) 309–319

needs, but also to use it as a control parameter for inter-nal processes or even as a tool for some of their requiredprocesses. Then the identity itself should be a processopen to “becoming”, and not a fixed point.

Then, the question is: what is the role of action inthe environment within a theory of autonomy? A broadagential autonomy implies that the organization of thesystem causes the processes exerted on the environment,whereas those of the environment towards the systemare monitored according to internally defined needs.To justify this we would like to distinguish betweenconstitutive processes, which produce the identity andlargely delimit what the system is, from interactive pro-cesses, which are not only side effects of the former,but crucial to maintain the identity of the system, withthe specific function of controlling the interaction withthe environment. We may picture those two, consti-tutive and interactive, aspects of autonomy as actingin different temporal scales, being the first faster andmore fundamental than the second, although both areequally required. It might be clarifying to think of theexample of the active transport of ions across the mem-brane, required to prevent osmotic crisis. This interaction(a form of “work”, as it carries ions against gradient)requires an internal sub-organization of different chainedreactions. The cell can be maintained due to ion trans-port interaction, but this can be realized because there isa pre-existent internal system.

Therefore, we are proposing that autonomy requiresmore than self-organization or self-maintenance. In asimilar way, Bickhard (2004) noticed the need of an“infrastructure” in the system, understood as an internalmechanism able to organize and channel energy flows forthe system’s self-maintenance. This subsystem should besome form of modulation of the self-maintaining pro-cesses themselves.

A consequence of this is that autonomy requires func-tional organization. As it is known, the Kantian approachto organisms describes them as natural purposes and,in his view, this intrinsic teleology makes it difficultto explain them in mechanistic terms. In opposition tothis, Maturana and Varela proposed to explain organ-isms through a circular organization very similar to theKantian, but they consider organisms as “autopoieticmachines”, and explicitly aim to avoid teleology.5 Now,

5 However, at the end of his life, Varela (Varela and Weber, 2001)adopted a different position, closer to the Kantian approach.

System

ft

3

tlaesotmo

tesb

o

tcnTi(eeitc

d

nal relations involve the loss of the autonomy of theincorporated system. In multicellular organization, thenew constituent parts, previously independent, do not

6 In particular, she claims that representational re-description mech-anisms are necessary to produce deliberate self-control.

7 By a wider functional universe we mean an increase in the numberand variety of the interactive processes for ensuring the maintenanceof the system, along with an increase in the number of hierarchicalregulatory controls.


unction is required by even the simplest satisfactoryheory of life or autonomy.

. Levels, degrees and domains of autonomy

A key conceptual aspect of autonomy is the distinc-ion of its different kinds and degrees, but it has beenargely neglected. Maturana and Varela explained livingnd cognitive phenomenology starting from the autopoi-tic organization, and even if they acknowledged that thetructures changed in history, they considered that therganization itself was not affected by history. Never-heless, we think that an adequate theory of autonomy

ust try to address degrees of autonomy; i.e. the aspectf becoming (more) autonomous.

One of the difficulties for a comparative study ofhe domains in which autonomy appears (with differ-nt degrees) is that all are characterized according to theame type of circular organization. For example, at theiological level the autopoietic organization is:

“a network of processes of production (transfor-mation and destruction) of components which: (i)through their interactions and transformations con-tinuously regenerate and realize the network ofprocesses (relations) that produced them; and (ii) con-stitute it (the machine) as a concrete unity in spacein which they (the components) exist by specifyingthe topological domain of its realization as such anetwork.“(Maturana and Varela, 1973, p. 78).

And cognitive autonomy is closure of the dynamicsf the nervous system:

“two main specifying characteristics of cognitive selfare given by the nervous system’s operational closurewhich: a) produces invariant sensory-motor patternsb) and specifies the organism as a mobile unit inspace” (Bourgine and Varela, 1991, p.).

With this strategy it is difficult to connect thesewo different domains of autonomy in terms of theomplexity of the structures required to present the phe-omenological properties associated to each of them.he first one takes into account the production of (chem-

cal) components, and the second of the sensorimotoror neural) patterns. From these definitions no differ-nce can be drawn related to the structural properties ofach one, only the similar circular logic of each systems enhanced. This suggests that the differences between
hese two kinds of systems are not relevant in what con-erns autonomy, but we think they might be crucial.
If the autonomous system is an agent whose identityepends on constitutive and interactive processes, struc-

s 91 (2008) 309–319 313

tural changes leading to a greater independence fromthe environment or to greater control over how it influ-ences the agent become a clue to identify increases inautonomy. This is the idea behind some attempted char-acterizations. For example, in Cariani’s view (1998), asystem acts more autonomously if it depends more oninternal processes than on external inputs. In a similarway, Boden (1996) considers that:

“an individual’s autonomy is the greater, the more itsbehavior is directed by self-generated (and idiosyn-cratic) inner mechanisms, nicely responsive to thespecific problem-situation, yet reflexively modifiableby wider concerns.” (p. 102).

Boden’s point of view here is that the degree ofautonomy is linked to the system’s capacity for self-modification of behavior producing mechanisms6 (seealso Boden, this issue).

The problem with these approaches lies on thedifficulty to determine what higher degrees of self-modification are. It is plain that many unicellularorganisms already show high capacities for self-modification, even higher than those of complexmulticellulars in terms of self-repair, morphological self-modification and alike. However, autonomy is a capacitythat may increase in history, in ontogenetic and phyloge-netic terms. This becoming might be seen as an access toa wider interactive functional universe,7 a higher degreeof autonomy appears to be related with the creation ofmore complex, hierarchically organized, functional con-straints on the environment. In evolutionary terms theprocess of becoming more autonomous sometimes con-sists in the integration of already existing autonomoussystems via appropriate controls.8 This process mayinvolve that what previously were interactive processesamong autonomous systems result in internal or consti-tutive processes of a more encompassing system. In thecase of the origins of eukaryotic cells, the new inter-

8 Complex interactions among autonomous systems can also be seenas trade-offs of mutual benefits (Christensen and Bickhard, 2002).When collective benefits are limited, the autonomy of the more encom-passing system is weak, whereas parts retain significant autonomy. Seealso Buss (1987).

System

association between the evolution of highly integratedand complex bodies and the evolution of cognitive auton-omy (Moreno and Lasa, 2003; Moreno and Etxeberria,

11 Self-organization alone cannot explain this process. An explanationfor it can go along the following lines: “(. . .) whilst ‘self-organization’is celebrated for its capacity to generate global patterns it has signif-icant limitations as a means of resolving the problems presented byintegration pressure. The most important of these are slow action and


loose metabolic autonomy, but their interactive processesbecome heavily constrained. In what respects coloniesand societies the cohesion of the collective has differentgrounds in each of them. The first depends mainly ona self-organizing dynamics producing an encompassingagential capacity that raises the adaptability that isolatedparts would have; occasionally they may almost appearas an integrated multicellular organism.9 But colonyautonomy does not become a lot more complex. In thecase of animal societies, individual organisms remainautonomous to a large extent. This is probably linkedto the fact that societies are constituted by organismsendowed with cognitive systems, which show more intri-cate forms of agency than the integrants of colonies.Societies are organized according to complex hierarchiesof regulatory controls that provide suitable environmentsfor complex forms of agency, like cognitive communi-cation. However, although in certain cases it may appearthat the society as a whole behaves as an agent, theidentity causally grounding the actions in these kindsof systems is by far more complex at the level of theindividuals than at the level of the society.

The case of multicellular organisms is quite different.Here new and more complex forms of agency appearin the evolution of animals, grounded on the stabiliza-tion of self-regulated functions through mechanisms ofinternalization able to protect them against, environ-mental perturbations. The way for evolution to achievehighly integrated multicellular systems seems to requiresomething more than forms of self-organization amongan increasing number of constitutive systems, like incolonies and certain societies. Multicellular organisms,specially in the case of animals, require the creationof complex regulatory mechanisms in order to gen-erate an integrated functional unit from the relationsamong the constitutive cells. All these processes con-tribute to an increase of autonomy, at least they generatephenomenologies that only metaphorically adjust tothe autopoietic model of autonomy: while they are
still cases of self-production, a lot more than that canbe said about them. In evolution other processes ofautonomization10 have been proposed to illustrate some
9 This is the case, for example, of the so-called “magnetotactic mul-ticellular prokaryote”, a bacterial aggregate that exhibits an unusual“ping-pong” motility in magnetic fields (Keim et al., 2004).10 “Increasing autonomy is defined as the evolutionary shift in the evo-

lutionary system/environment relationship, so that the direct influencesof the environment are gradually reduced and stability and flexibilityof self-referential, intrinsic functions within the system is generated”Rosslenbroich, 2006, p. 61. The term autonomization was used bySchmalhausen (1949).

s 91 (2008) 309–319

form of progress towards more independence from theenvironment (Rosslenbroich, 2005, 2006). For instance,the extracellular matrix appearing common in the devel-opment of metazoan, which appeared early in evolution,allows intracellular conditions to regulate and protectinternal cells from the external environment. This inde-pendence is relative because it is built at the sametime that “many interconnections with and dependenciesupon” the environment are retained and perhaps elabo-rated. Some of the elements are: spatial separation fromthe environment (membrane, walls, intergumens etc.),establishment of homeostatic functions, internalizationof morphological structures or function from an externalposition.

The creation of regulatory mechanisms requires ahierarchical organization, in which the degree of auton-omy of the constituents is restrained. As we have pointedout, the way for evolution to achieve highly complexbiological systems does not seem to be to increase thenumber and variety of the constituent parts organized ina distributed manner, but to generate new, nested formsof regulatory control.11 For example, the increasing pro-cess of autonomization in the evolution of vertebratesgoes together with the fact that their metabolic organi-zation is fully and precisely controlled by their brain.Their characteristic agency has been made possible bythe development of the nervous system, which evolvedas a powerful regulatory mechanism to control and inte-grate complex underlying processes. There is a strong

poor targeting capacity. Precisely because achieving the global statedepends on propagating state changes through many local interactionsthe time taken to achieve the final state can be long, and increaseswith the size of the system. Moreover, since there is no regulation ofglobal state, the ability of the system to find the appropriate collectivepattern depends on the fidelity of these interactions. Here there is atension: if the self-organization process is robust against variations inspecific conditions the process will be reliable, but it will be difficultfor the system to generate multiple finely differentiated global states.Alternatively, if the dynamics are sensitive to specific conditions itwill be easy for the system to generate multiple finely differentiatedglobal states, but difficult to reliably reach a specific state. (. . .) Con-sequently the most effective means for achieving the type of globalcoherence required for functional complexity is through regulation,including feedback mechanisms and instructive signals operating atboth local and larger scales”. (Christensen, 2007).

System

2aaacamc

4

sotca

ttdihIllmaietd

iahb1oa

uoeusc

“t


005). In other words, systems with higher degrees ofutonomy show an increase in the capacity to creatend/or take over complex and larger environmental inter-ctions, because of a more intricate organization of theironstitutive identity. Their autonomy is also based oncircular, recursive organization, but this also includesany hierarchical levels and many embedded regulatory

ontrols.

. Autonomy and the artificial

As already said, autonomy is a challenge for artificialystems: in the ontological sense, because the autonomyf artificial systems is limited; in the epistemological,he question is whether we can learn about autonomy byonstructing artificial models, and what kind of modelsre required.

The challenge is the creation of systems that are, athe same time, artificial and not fully designed. Thoughhe goal to create a true autonomous system should beifferentiated from the epistemological aim of construct-ng artificial systems as models, these two objectivesave been often linked in the sciences of the artificial.n the field of Artificial life, for example, the study ofiving organisms was faced “by attempting to synthesizeife-like behaviors within computers and other artificial

edia” (Langton, 1989, p. 1). Here the very concept ofutonomy becomes both the end and the way; if a systems made by strict design, it can hardly be autonomous:ither its actions would not reconfigure its identity, or, ifhey did, the system would not be anymore the result ofirect design.

As autonomy implies that the identity of the systems self-created, a hypothetical artificial production of anutonomous system would require an indirect form ofuman action, so that by partially “get(ting) the humaneing out of the loop” as Langton described (Boden,996; Risan, 1997), some of the human creativity turnsut to be externalized to the dynamical behavior of thertifact itself.

The originality of this project stems from a bottom-p, “emergent” methodology and the mimesis of somef the natural ways to generate complex processes (i.e.,
volution, development or learning). In computer sim-lations, the result does not have to be – at least in itspecific form – analytically inferred and it must be moreomplex12 than what has generated it, in the sense that
12 In the sense of less trivial, because it can be – and usually it is – asimplified” pattern with respect of the low level interactions that ledo it.

s 91 (2008) 309–319 315

the dynamic unfolding of the simulation is far richerthan the simple aggregation of the local rules that gen-erate it. The design of the model should be simple andsimilar to the conditions by which natural processes giverise to something more complex. With this aim, a set oftechniques, known under the common terms of “emer-gent computation” (Forrest, 1991), “artificial evolution”(Harvey et al., 2005), etc. have been developed to bringinto existence (along with progresses in hardware) anexplosion of computational “lifelike creatures”.

Two broad kinds of techniques have been developedto study living phenomenology using artificial systems.One follows a part/whole strategy aiming to replicate thecomplexity of the system taking into account the relationof constituents with the totalities they form. In general,the method used is bottom-up and tries to study the self-organizing properties of the system. Examples of thiskind of research line are the formation of chemical self-maintaining networks, protocells, structures in neuralnetworks, ant colonies, the organization of the immunesystem or of robot societies. The main conceptual andtechnical problem of this approach is how to capturethe circular causality that characterizes autonomy, whichshould be both top-down and bottom-up, so that themacroscopic behavior emerges from microscopic localrules of simple parts, whereas the macro emergent effectsfeed back to modify and control the bottom level. Thereare already good models that capture some of the cir-cularity and self-organizing properties characteristic ofautonomous systems (Varela et al., 1974; McMullin andVarela, 1997; Fontana, 1992). The challenge now lies ondeveloping models in which a richer repertoire of func-tional behavior is integrated into the model and emergentfrom its local interactions, together with the explorationof spontaneous (i.e., not externally imposed) formationof a hierarchical regulatory organization.

The other strategy is artificial evolution, which triesto substitute the human design of artificial agents byan evolutionary process. In what respects the evolu-tion of autonomous systems, most of the first modelsof artificial evolution relied on a relatively adaptation-ist version of Darwinian evolution, partly obliged bythe use of fitness functions to guide the search pro-cess, and mainly because the evolving structures wererepresented as idealized genotypes, instead of beingself-reproducing autopoietic systems (Griesemer, 2000;Etxeberria, 2000, 2004). The role attributed to artifi-cial evolution regarding the modeling of autonomous
systems has been generally limited to optimize the sys-tem towards a unique desired functionality (which was,in addition, generally externally defined). As a result,the system was generally very limited on its behav-

System
ioral repertoire and, most importantly, its behavior wasuncoupled to internal (self-maintaining or stability) con-ditions. Recent attempts to overcome these limitations(in particular those focused on selection for internal sta-bility coupled to adaptive behavior (Di Paolo, 2003, thisissue) might be able to generate more complex forms ofautonomous organization, making room for functionaldiversity and the necessary hierarchical regulatory orga-nization required to manage it. It is within this contextwhere the previously raised questions related to the inter-play between levels, degrees and domains of autonomycould be addressed. But evolution itself, that is to say,the problem of the role of autonomy in evolutionaryprocesses, is still largely neglected, and the proposalof “natural drift” as the main evolutionary phenomenon(Maturana and Mpodozis, 2000) has not been pursuedfurther in artificial systems.

Until very recently, most of the models more directlyaiming to reproduce the constitutive organization ofautonomy ignored or neglected thermodynamic and/orenergetic constraints. However, since autonomous sys-tems cannot be but far from equilibrium dissipativeorganizations (where the flow of matter and energyacross the system makes interactive and constitutiveprocesses inherently interdependent), the abstraction ofthese aspects in the design of models hides the stronginterconnection between the interactive and the con-stitutive dimensions of real autonomous systems. Thislimitation is currently being overcome in some recentwork on more realistically simulated chemical networksthat could lead to minimal autonomous systems (Daleyet al., 2002; Kauffman, 2003; Olasagasti et al., 2007;Mavelli and Ruiz-Mirazo, 2007), although still in a verypreliminary stage of research.

Another, complementary, way to study minimalautonomy is “synthetic biology”. Synthetic biology canbe considered as a redefinition and expansion of biotech-nology, with the ultimate goals of being able to buildengineered biological system (Benner and Sismour,2005). Thus, it goes further than classical genetic engi-neering and seeks the complete fabrication of livingbeings starting from the same (or very similar) natu-ral materials. On these lines there are currently differentresearch programs under development that involve thefabrication of some sort of minimal artificial organism orprotocell (Sole et al., 2007). On the one hand, followinga ‘top-down’ strategy, different researchers are trying tofind out the simplest form of a living cell by modifying
extant unicellular organisms with genetic engineeringtechniques. The aim is to find an artificial cell with theminimal genome able to sustain their most basic vitalfunctions (Hutchinson et al., 1999; Cho et al., 1999;
s 91 (2008) 309–319

Luisi et al., 2002). On the other hand, from a ‘bottom-up’ approach, artificial life in vitro (Szostak et al., 2001;Pohorille and Deamer, 2002) aims to synthesize mini-mal systems with capacity for self-maintenance and/orreproduction, starting from the most basic molecularcomponents.

Finally, we have to mention the important researchline of biologically inspired robotics (Steels and Brooks,1995; Beer, 1997; Webb, 2001). This research is mainlyfocused on self-organized sensorimotor interactions,often neglecting the relationship between interactive andconstitutive aspects of autonomy by focusing exclu-sively on the former. However in the complex interactiveautonomy the guiding motives for behavior cannot bereduced to self-maintenance or survival, and that is whysome researchers have proposed that the sensorimotordomain itself may generate autonomous forms of con-stitutive organization (Di Paolo, 2003; Barandiaran andMoreno, 2006). Particularly important for research inthis direction is how the notions of value, intentionalityand emotions relate to the concept of autonomy and itsdynamic organization (Di Paolo and Iizuke, this issue;Ziemke, this issue).

To sum up, current approaches to artificial model-ing are largely biased by a partial view of autonomy.As we have seen in Section 2, for a system to beautonomous, in its minimal sense, it has to fulfill cer-tain requirements. And in Section 3, we have developedthis concept, showing that it allows to deal with the vari-eties and degrees that biological evolution has unfolded.These requirements are more demanding than many ofthe standard views on autonomy developed by the the-ory of autopoiesis or by current biologically inspiredrobotics. On the one hand, most of the popular researchmethodologies for modelling autonomy (i.e., evolution-ary robotics, dynamical systems approach, etc.) arealmost exclusively focused on the study of the interactivedimension of autonomy, neglecting the constitutive one.On the other hand, the – relatively few – models deal-ing with the constitutive dimension tend to ignore theinteractive aspect. This is quite understandable, since itis easier to study limited aspects of autonomy than itswhole complexity. However, we consider that furtheradvances will require the development of new models(some of which are already on their way) that take intoaccount both the constitutive and the interactive dimen-sion of autonomy together, as some of the key featuresof autonomy cannot be understood but as arising from
the intimate coupling between the two mentioned dimen-sions. Ruiz-Mirazo and Mavelli (this issue) and Ikegamiand Suzuki (this issue) constitute preliminary attemptsto achieve this interconnection between constitutive and

System

iwiil

5

aeattaot(Itnceocse

dewtahplieHpttd

A

9tHct


nteractive aspects of autonomy in the metabolic domain,hile Ziemke (this issue) and Di Paolo and Iizuke (this

ssue) show complementary approaches to deal withnteractive and constitutive aspects at the sensorimotorevel.

. Conclusions

We have taken into account that the theory ofutopoiesis started a research line on autonomy thatxtends classical philosophical approaches in thatutonomous organization is not only behind the capacityo behave autonomously, but it also defines the sys-em ontologically. This ontological account assimilatesutonomy to the kind of systems that are alive. However,ur approach to autonomy deviates from autopoiesis inhat we try to discuss and contribute towards a materialand not merely organizational) and historical concept.n this sense, we have tried to explain why we considerhat autonomy should not be considered only in inter-al constitutive terms, but that both the interactive andonstitutive aspects stemming from autonomy should bequally taken into account. The main consequences ofur focus are the need to explain function and hierarchi-al integration of parts within the theory of autonomy,tarting from self-organization but in such a way that thexplanation of autonomy is not exhausted by it.

We think that this approach makes it possible toevelop better suggestions for work on artificial mod-ls. By attempting to (re)produce autonomous systems,e can at the same time learn how and why they are

he way they are. As the fields of Cybernetics beforend both Artificial Intelligence and Artificial Life laterave showed, progress in our understanding about theroperty of autonomy in general and the autonomy ofiving and cognitive systems in particular requires anntensive use of computational simulations, robotic mod-ls and even synthetic biochemistry and biotechnology.owever, the main goal of this activity should not be toroduce artificial autonomous systems, but to get a bet-er understanding of the role of autonomy for life andhe varieties of its organization and phenomenologicaliversity.

cknowledgements

Funding for this work was provided by grants/UPV00003.230-15840/2004 from the University of
he Basque Country-Euskal Herriko Unibertsitatea, andUM2005-02449 from the Ministry of Science and Edu-
ation (MEC) and Feder funds from the E.C. We thankhe valuable suggestions of two anonymous referees and

s 91 (2008) 309–319 317

of the editors of this issue (X. Barandiaran and K. Ruiz-Mirazo) to a previous version.

References

Atkins, P.W., 1984. The Second Law. Freeman, New York.Barandiaran, X., Moreno, A., 2006. On what makes certain dynamical

systems cognitive. a minimally cognitive organization program. J.Adapt. Behavior 14 (2), 171–185.

Beer, R., 1997. The dynamics of adaptive behavior: A research pro-gram. Robot. Auton. Syst. 20 (2–4), 257–289.

Benner, S.A., Sismour, A.M., 2005. Synthetic biology. Nat. Rev. Genet.6, 533–543.

Bickhard, M.H., 2004. The dynamic emergence of representation. In:Clapin, H., Staines, P., Slezak, P. (Eds.), Representation in Mind:New Approaches to Mental Representation. Elsevier, Amsterdam,pp. 71–90.

Boden, M., 1996. Autonomy and artificiality. In: Boden, M. (Ed.), ThePhilosophy of Artificial Life. Oxford University Press, Oxford, pp.95–108.

Boden, M., this issue. Autonomy: What is it?Boogerd, F., Bruggeman, F., Hofmeyr, J., Westerhof, H. (Eds.), 2007.

Systems Biology: Philosophical Foundations. Elsevier, Amster-dam.

Bourgine, P., Varela, F.J., 1991. Towards a practice of autonomoussystems. In: Bourgine P., Varela, F.J. (Eds.), Towards a Practice ofAutonomous Systems. Proceedings of the First European Confer-ence on Artificial Life. MIT Press, Cambridge, MA. pp. xi–xvii.

Buss, L.W., 1987. The Evolution of Individuality. Princeton UniversityPress, Princeton, NJ.

Cariani, P., 1998. Epistemic autonomy through adaptive sensing. In:Proceedings of the 1998 IEEE ISIC/CRA/ISAS Joint Conference,Gaithersburg, MD, September 14–16, 1998, pp. 718–723.

Cho, M.K., Magnus, D., Caplan, A.L., McGee, D., The Ethics ofGenomics Group, 1999. Ethical considerations in synthesizing aminimal genome. Science 286 (5447), 2087–2090.

Christensen, W., 2007. The evolutionary origins of volition. In:Spurrett, D., Kincaid, H., Ross, D., Stephens, L. (Eds.), DistributedCognition and the Will: Individual Volition and Social Context.MIT Press, Cambridge MA.

Christensen, W., Bickhard, M., 2002. The process dynamics of norma-tive function. Monist 85 (1), 3–28.

Daley, A.J., Girvin, A., Kauffman, S.A., Wills, P.R., Yamins, D., 2002.Simulation of chemical autonomous agents. Z. Phys. Chem. 216,41–49.

Di Paolo, E.A., 2003. Organismically-inspired robotics: homeostaticadaptation and natural teleology beyond the closed sensorimo-tor loop. In: Murase, K., Asakura, T. (Eds.), Dynamical SystemsApproach to Embodiment and Sociality. Advanced KnowledgeInternational, Adelaide, Australia, pp. 19–42.

Di Paolo, E.A., 2005. Autopoiesis, adaptivity, teleology, agency. Phe-nomenology and the Cognitive Sciences 4 (4), 429–452.

Di Paolo, E.A. this issue. How (not) to model autonomous behaviour.Etxeberria, A., 2000. Artificial evolution: creativity and the possible.

In: Bedau, M., McCaskill, J., Norman Packard, N., Rasmussen, S.(Eds.), Artificial Life V. MIT Press, Cambridge, MA, pp. 555–562.

Etxeberria, A., 2004. Autopoiesis and natural drift: genetic infor-mation, reproduction, and evolution revisited. Artif. Life 10 (3),347–360.

Etxeberria, A. & Umerez, J. (2006). Organizacion y organismo en laBiologıa Teorica ¿Vuelta al organicismo? Ludus Vitalis 26.

System
Etxeberria, A., Moreno, A., Umerez, J. (Eds.), 2000. Special issue onthe contribution of artificial life and the sciences of complexity tothe understanding of autonomous systems. CCAI: Commun. Cogn.Artif. Intell. 17 (3/4), 111–254.

Fontana, W., 1992. Algorithmic chemistry. In: Langton, C.G., Taylor,C., Farmer, J.D., Rasmussen, S. (Eds.), Artificial Life II. Addison-Wesley, Redwood City, MA, pp. 159–209.

Forrest, S. (Ed.), 1991. Self-Organizing, Collective, and CooperativePhenomena in Natural and Artificial Computing Networks. MITPress, Cambridge, MA.

Griesemer, J., 2000. Reproduction and the reduction of genetics. In:Beurton, P., Falk, R., Rheinberger, H.-J. (Eds.), The Concept ofthe Gene in Development and Evolution. Historical and epistemo-logical perspectives. Cambridge University Press, Cambridge, pp.240–285.

Harvey, I., Di Paolo, E.A., Tuci, E., Wood, R., Quinn, M., 2005. Evolu-tionary robotics: a new scientific tool for studying cognition. Artif.Life 11, 79–98.

Hayles, N.K., 1999. How we became posthuman: virtual bodies incybernetics. In: Literature and Informatics. The University ofChicago Press, Chicago.

Hutchinson III, C.A., Peterson, S.N., Gill, S.R., Cline, R.T., White,O., Fraser, C.M., Smith, H.O., Venter, J.C., 1999. Global transpo-son mutagenesis and a minimal mycoplasma genome. Science 286(5447), 2165–2169.

Ikegami, T., Suzuki, K., this issue. From homeostatic to homeodynamicself.

Jonas, H., 2001. The phenomenon of life. Toward a philosophicalbiology. Northwestern University Press, Chicago, ILL.

Kauffman, S.A., 2003. Molecular autonomous agents. Phil. Trans. R.Soc. London A 361, 1089–1099.

Keim, C.N., Martins, J.L., Abreu, F., Rosado, A.S., Lins de Barros,H., Borojevic, R., Lins, U., Farina, M., 2004. Multicellular lifecycle of magnetotactic prokaryotes. FEMS Microbiol. Lett. 240(2), 203–208.

Kauffman, S., 2000. Investigations. Oxford University Press, Oxford.Keller, E.F., 2007. The disappearance of function from ‘self-organizing

systems’. In: Boogerd, F., Bruggeman, F., Hofmeyr, J.-H., West-erhoff, H.V. (Eds.), Systems Biology. Philosophical Foundations.Elsevier, Amsterdam.

Kitano, H., 2002. Computational Systems Biology. Nature 420,206–210.

Langton, C.G., 1989. Artificial life. In: Langton, C. (Ed.), ArtificialLife I. Addison Wesley, Redwood City, CA, pp. 1–47.

Luisi, P.L., Oberholzer, T., Lazcano, A., 2002. The notion of aDNA minimal cell: A general discourse and some guidelinesfor an experimental approach. Helvetica Chim. Acta 85, 1759–1777.

Maturana, H.R., Varela, F.J., 1973, De maquinas y Seres Vivos.Autopoiesis: La organizacion de lo Vivo. Editorial Universitaria.Santiago (1994, 3rd edition including new prefaces by each author).

Maturana, H., Varela, F., 1980. Autopoiesis and Cognition: the Real-ization of the Living. Reidel, Dordecht.

Maturana, H., Varela, F., 1984. El arbol del conocimiento. EditorialUniversitaria, Santiago de Chile.

Mavelli, F., Ruiz-Mirazo, K., 2007. Stochastic simulations of minimalself-reproducing cellular systems. Philo. Trans. Roy. Soc. London
B. 362, 1789–1802.
McMullin, B., Varela, F., 1997. Rediscovering ComputationalAutopoiesis. In: Husbands, P., Harvey, I., (Eds.), Fourth EuropeanConference on Artificial Life. MIT Press, Cambridge, MA, pp.38–47.

s 91 (2008) 309–319

Moreno, A., Lasa, A., 2003. From basic cognition to early mind. Evol.Cogn. 9 (1), 12–30.

Moreno, A., Etxeberria, A., 2005. Agency in natural and artificialsystems. Artif. Life 11 (1–2), 161–175.

Maturana, H.R., Mpodozis, J., 2000. The origin of species bymeans of natural drift. Revista Chilena de Historia Natural 73,261–310.

Nolfi, S., Floreano, D., 2000. Evolutionary Robotics: The Biology,Intelligence, and Technology of Self-Organizing Machines. MITPress/Bradford Books, Cambridge, MA.

Olasagasti, F., Moreno, A., Pereto, J., Moran, F., 2007. Energeticallyplausible model of a self-maintaining protocellular system. BullMath Biol. 69 (4), 1423–1445.

O’Malley, M.A., Dupre, J., 2005. Fundamental issues in systems biol-ogy. BioEssays 27, 1270–1276.

Pattee, H.H., 1972. Laws and constraints, symbols and languages.In: Waddington, C.H. (Ed.), Towards a Theoretical Biology4, Essays. Edinburgh University Press, Edinburgh, pp. 248–258.

Pattee, H.H., 1973. The physical basis and origin of hierarchical con-trol. In: Pattee, H.H. (Ed.), Hierarchy Theory: The Challenge fromComplex Systems. Braziller, New York, pp. 71–108.

Pattee, H.H., 2001. The physics of symbols: bridging the epistemiccut. BioSystems 60 (1/3), 5–21.

Pohorille, A., Deamer, D.W., 2002. Artificial cells: prospects forbiotechnology. Trends Biotechnol. 20 (3), 123–128.

Risan, L., 1997. Why are there so few biologists there? Artificial lifeas a theoretical biology or artistry. In: Husbands, P., Harvey, I.,(Eds.), Fourth European Conference on Artificial Life, MIT Press,Cambridge, MA, pp. 28–35.

Rosslenbroich, B., 2005. The evolution of multicellularity in ani-mals as a shift in biological autonomy. Theory Biosci. 123,243–262.

Rosslenbroich, B., 2006. The notion of progress in evolutionary biol-ogy. The unresolved problem and an empirical suggestion. Biol.Philo. 21, 41–70.

Ruiz-Mirazo, K. Mavelli, F., this issue. On the way towards ‘basicautonomous agents’: stochastic simulations of minimal lipid-peptide cells.

Schmalhausen, I., 1949. Factors of Evolution. The University ofChicago Press, Chicago.

Science, 2002. Special issue on systems biology. Science 295,1661–1682.

Simon, H.A., 1996. [1969,3rd ed.] The Sciences of the Artificial. MITPress, Cambridge, MA.

Sole, R.V., Munteanu, A., Rodriguez-Caso, C., Macıa, J., 2007. Syn-thetic Protocell Biology: from reproduction to computation. Philo.Trans. Roy. Soc. London B 362, 1789–1802.

Steels, L., Brooks, R.A. (Eds.), 1995. The Artificial Life Route to Arti-ficial Intelligence: Building Embodied Situated Agents. LawrenceErlbaum, Hillsdale, NJ.

Szostak, J.W., Bartel, D.P., Luisi, P.L., 2001. Synthesizing life. Nature409, 387–390.

Turner, J.S., 2000. The Extended Organism: the Physiology of Animal-built Structures. Harvard University Press, Cambridge, MA.

Varela, F.J., Maturana, H., Uribe, R., 1974. Autopoiesis: the organiza-tion of living systems, its characterization and a model. BioSystems
5, 187–196.
Van de Vijver, G., Van Speybroeck, L., Vandevyvere, W., 2003. Reflect-ing on complexity of biological systems. Kant and beyond? ActaBiotheoretica 51 (2), 101–140.

Weaver, W., 1948. Science and complexity. Am. Sci. 36 (4), 536–544.

System

W

W


ebb, B., 2001. Can robots make good models of biologicalbehaviour? Behav. Brain Sci. 24, 1033–1050.

imsatt, W.C., 2007. On building reliable pictures with unreliabledata: An evolutionary and developmental coda for the new systemsbiology? In: Boorgerd, F., Bruggerman, F., Hofmeyr, J.-H., West-

s 91 (2008) 309–319 319

erhoff, H.V. (Eds.), Systems Biology. Philosophical Foundations.Elsevier, Amsterdam, pp. 103–120.

Ziemke, T., this issue. The role of emotion on biological and roboticautonomy.

the autonomy of biological individuals and artificial models

Documents